Add artifact-labeling ability from "An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training"

[ericearl]

What would you like to see added in this software?

The labeling ability of specific MRI artifacts from "An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training".

• https://doi.org/10.1016/j.media.2023.103033
• https://github.com/Queen-Square-Analytics/automatic-quality-control

Do you have any interest in helping implement the feature?

Yes, but I would need guidance

Additional information / screenshots

[oesteban]

It looks that the weights are not shared, is that correct?

[ericearl]

@oesteban I'm not sure. It looks like some hard-coded parameters are present in specific artefact noise generators:

• https://github.com/Queen-Square-Analytics/automatic-quality-control/blob/main/src/qcs/artefacts/ghosting.py

And this seems to default to ResNet101 weights>

• https://github.com/Queen-Square-Analytics/automatic-quality-control/blob/main/src/qcs/feature_extraction/deep.py#L42

But otherwise, yes, I don't readily see model weights. Perhaps a conversation could be started with the authors?

[oesteban]

Perhaps a conversation could be started with the authors?

Sure, maybe they are actually interested in delivering their model through MRIQC

Perhaps off-topic, but this could be interesting too https://arxiv.org/pdf/2509.24888